The Role of Factor XIa (FXIa) Catalytic Domain Exosite Residues in Substrate Catalysis and Inhibition by the Kunitz Protease Inhibitor Domain of Protease Nexin 2*

To select residues in coagulation factor XIa (FXIa) potentially important for substrate and inhibitor interactions, we examined the crystal structure of the complex between the catalytic domain of FXIa and the Kunitz protease inhibitor (KPI) domain of a physiologically relevant FXIa inhibitor, protease nexin 2 (PN2). Six FXIa catalytic domain residues (Glu98, Tyr143, Ile151, Arg3704, Lys192, and Tyr5901) were subjected to mutational analysis to investigate the molecular interactions between FXIa and the small synthetic substrate (S-2366), the macromolecular substrate (factor IX (FIX)) and inhibitor PN2KPI. Analysis of all six Ala mutants demonstrated normal Km values for S-2366 hydrolysis, indicating normal substrate binding compared with plasma FXIa; however, all except E98A and K192A had impaired values of kcat for S-2366 hydrolysis. All six Ala mutants displayed deficient kcat values for FIX hydrolysis, and all were inhibited by PN2KPI with normal values of Ki except for K192A, and Y5901A, which displayed increased values of Ki. The integrity of the S1 binding site residue, Asp189, utilizing p-aminobenzamidine, was intact for all FXIa mutants. Thus, whereas all six residues are essential for catalysis of the macromolecular substrate (FIX), only four (Tyr143, Ile151, Arg3704, and Tyr5901) are important for S-2366 hydrolysis; Glu98 and Lys192 are essential for FIX but not S-2366 hydrolysis; and Lys192 and Tyr5901 are required for both inhibitor and macromolecular substrate interactions.

Factor XI (FXI) 3 is a 160-kDa homodimeric protein (1) that circulates in human plasma in complex with high molecular weight kininogen at a concentration of 30 nM (4 -6 g/ml) (1,2). Deficiency of FXI results in a bleeding diathesis referred to as hemophilia C that is most common in individuals of Askenazi Jewish descent (3,4). FXI is activated to FXIa by cleavage of the scissile bond between Arg 369 and Ile 370 by factor XIIa (FXIIa) or thrombin or by autoactivation in the presence of a negatively charged surface (1,5,6). Upon activation, each of the identical subunits contains a 50-kDa heavy chain and a 30-kDa light chain. The NH 2 -terminal heavy chain is composed of four tandem repeat sequences called Apple domains (Apple 1 to Apple 4), and the COOH-terminal light chain contains the catalytic triad residues His 413 , Asp 462 , and Ser 557 (His 57 , Asp 102 , and Ser 195 ; chymotrypsin numbering system, which will be used throughout this paper) (7). During cleavage of FXI, a new NH 2terminal sequence, Ile 16 -Val 17 -Gly 18 -Gly 19 , is formed, which is characteristic of serine proteases. The NH 2 -terminal Ile 16 inserts into the protease domain of FXIa, and the NH 2 group forms a salt bridge with the COOH group of Asp 194 . This salt bridge is a defining feature during the formation of FXIa (7).
FIX is the natural macromolecular substrate of FXIa. The Ca 2ϩ -dependent activation of FIX by FXIa (8,9) requires the exposure of a substrate-binding site within the Apple 2 and/or Apple 3 domain of FXIa and the ␥-carboxyglutamic acid domain of FIX, as well as an extended, macromolecular substrate-binding exosite in the protease domain of FXIa (10 -14). The activation of FIX to FIXa␤ involves two cleavages by FXIa, one after Arg 145 and another after Arg 180 , which releases an 11-kDa activation peptide (8,9,15). FIXa␤ is also produced by the tissue factor-factor VIIa complex (16).
Protease nexin 2 (PN2) is a Kunitz-type protease inhibitor (KPI) secreted by activated platelets (17)(18)(19) that has been shown to have high affinity and specificity for FXIa. The interaction between PN2 and FXIa has previously been shown to involve interactions that occur exclusively between the KPI domain of PN2 (PN2KPI) and the catalytic domain of FXIa (FXIac) (20). The isolated KPI domain and the FXIa catalytic domain have been co-crystallized, and their structure has been solved to a resolution of 2.6 Å (21). This structure combined with a mutational analysis of the KPI domain has been used to identify a number of residues within two loop structures (Loop 1 and Loop 2) within the KPI domain postulated to interact with corresponding residues within the catalytic domain of FXIa that are potentially important for both inhibitor and substrate interactions. We have therefore utilized this structural information ( Fig. 1) to examine the architecture of residues in close proximity to the catalytic triad and to select a number of residues within the catalytic domain of FXIa (Asp 98 34 and Tyr 35 ) of the KPI domain of PN2 (see Table 1). It should be noted that Arg 3704 (alternatively referred to as Arg 37D ), a residue unique to FXIa, is the fourth amino acid after residue 37 (chymotrypsin numbering) (7), residue 395 in mature FXI, or residue 76 of the catalytic domain of FXIa, whereas similarly Tyr 5901 (alternatively referred to as Tyr 59A ) is the first residue after residue 59. In the present work, we have made selected mutations at these identified exosite residues (i.e. excluding the active site), and examined the resulting enzymes (after activation to FXIa) in the hydrolysis of the peptide substrate S-2366, in the activation of the macromolecular substrate, FIX, and in the regulation of FXIa by PN2. The rationale for selecting these residues for mutational analysis includes the fact that Glu 98 is part of the 90s loop (residues 94 -100) of FXIa, a surface-exposed loop that varies in length and conformation among serine proteases. In FXIa, the 90s loop folds inward toward the catalytic triad residues and therefore may restrict the accessibility of substrates and inhibitors to this region. Residues Tyr 143 and Ile 151 are part of the autolysis loop (Tyr 143 -Thr 154 ) of FXIa. The basic residues within this loop have been previously shown to be important for FXIa serpin specificity (22). A surface-exposed residue unique to FXIa among serine proteases of blood coagulation and highly conserved among various species is Arg 3704 , which was therefore also chosen for mutational analysis. In this paper, in addition to examining the importance of these selected residues in both substrate hydrolysis and inhibitor (PN2KPI) recognition, we also examined the integrity of the S1 binding site residue, Asp 189 , utilizing the S1 site probe, p-aminobenzamidine (pAB). Our data demonstrate that the S1 site in all mutants is intact. Interestingly, all of the mutant proteins examined demonstrated significant decreases in the k cat values for macromolecular substrate (FIX) catalysis, and all (except for E98A and K192A) displayed significant decreases in the k cat values for small peptide hydrolysis, whereas the K m values for peptide hydrolysis were normal for all mutant enzymes. Although all of the residues chosen for mutational analysis were selected on the basis of interactions with PN2KPI, all mutants demonstrated abnormal macromolecular substrate recognition. Whereas mutations at four of these sites (Glu 98 , Tyr 143 , Ile 151 , and Arg 3704 ) resulted in normal values of K i for inhibition by PN2KPI, in contrast, mutations at the other two sites (Lys 192 and Tyr 5901 ) resulted in enzymes with impaired interactions with PN2KPI as well as macromolecular substrate catalysis.
Proteins-FXIIa, FXIa, and FXI purified from human plasma were purchased from Hematologic Technologies, Inc. (Essex Junction, VT). Kunitz-type protease inhibitor was purified from transfected Pichia pastoris as described previously (21). The monoclonal antibody 5F7 (directed against the A1 domain located within the heavy chain of FXI) was initially purified from the ascites fluids in a hybridoma cell line (23) and now is commercially available from Green Mountain Antibodies (Burlington, VT). Corn trypsin inhibitor (coupled to Affi-Gel) columns were purchased from Enzyme Research Laboratories (South Bend, IN).

FXI Mutant Constructs-
The cDNA for the full-length FXI sequence inserted in pJVCMV vector (a gift from Dr. David Gailani, Vanderbilt University, Nashville, TN) served as a template for the synthesis by PCR of the FXIa catalytic domain mutants. The mutations were introduced using a PCR-based site-directed mutagenesis kit (QuikChange TM ) using the appropriate mutagenic primers. The PCR products containing mutations were inserted into pJVCMV vector and were propagated in XL1-Blue bacteria. Each purified plasmid DNA was sequenced in the forward and reverse directions to verify that the appropriate mutation was incorporated.
Protein Expression in Human Embryonic Kidney (293) Cells and Purification-Human embryonic kidney cells (293-HEK) were transfected with 40 g of the pJVCMV vector containing inserts for FXI mutants and 2 g of pRSVneo vector (containing the gene that confers resistance to neomycin and allows the selection of positive clones) using Lipofectamine 2000. Positive clones were selected using Geneticin 418 (G-418) at a concentration of ϳ500 g/ml, and the expression levels were assessed by ELISA (described below). Cells were expanded in 2-liter roller bottles in DMEM containing 10% fetal bovine serum, penicillin/streptomycin, L-glutamine, and G-418 (ϳ150 g/ml final concentration) in a 5% CO 2 incubator, 37°C. After the cells reached confluence in the roller bottles, the medium was replaced with serum-free DMEM supplemented with penicillin/streptomycin (50 units/ml penicillin; 50 g/ml streptomycin), L-glutamine (0.3 mg/ml), G-418 (ϳ150 g/ml), insulintransferring selenium-A, soya bean trypsin inhibitor (10 g/ml), lima bean trypsin inhibitor (10 g/ml), and aprotinin (10 g/ml). Conditioned media were collected after 48 -72 h, centrifuged, and filtered through an acetate filter (0.45-m pore size) to remove any cell debris, were made 5 mM in EDTA and 5 mM in benzamidine to prevent any nonspecific protease cleavage of the protein, and were stored at Ϫ20°C until they were ready to be processed.
Expressed protein from cell supernatant was applied to the 5F7 monoclonal antibody affinity column equilibrated in 25 mM Tris-HCl, 100 mM NaCl, and 5 mM benzamidine, pH 7.4. The column was washed with equilibration buffer until the A 280 returned to base line. Adsorbed protein was eluted with 2 M potassium thiocyanate made in the equilibration buffer. The collected fractions were concentrated and dialyzed extensively against Tris-buffered saline, pH 7.4. The yield of mutant proteins ranged from 0.5 to 1.2 mg/liter of culture fluid. The purity of the fractions was assessed by SDS-PAGE before being pooled and concentrated to 0.25 ml for immediate use in experiments or for storage either at Ϫ80°C or in liquid nitrogen. As shown in supplemental Fig. 1, all FXI mutants were Ͼ95% pure as judged from SDS-PAGE (4 -15%) and like plasma FXI all migrated at 160 kDa nonreduced and at 80 kDa under reducing conditions (i.e. incubation with ␤-mercaptoethanol). The activated mutants like FXIa migrated as a single band nonreduced and as two bands of ϳ50 and ϳ30 kDa comprising the heavy and light chains, respectively.
Activation of FXI Mutants-FXI mutant proteins were activated overnight by FXIIa (10:1 molar ratio of FXI to FXIIa) at 37°C. The FXIIa was subsequently removed using a corn trypsin inhibitor column. Samples of the activated protein were analyzed by SDS-PAGE.
Active Site Titration-The active site concentrations of FXIa mutants were measured (24,25) utilizing the fluorogenic active site titrant 4-methylumbelliferyl 4-guanidinobenzoate hydrochloride monohydrate (MUGB) obtained from Sigma. The fluorescent product 4-methylubelliferone released by the hydrolysis of MUGB by FXIa WT or pFXIa was determined at excitation and emission wavelengths of 323 and 446 nm, respectively. FXIa WT (0 -200 nM) in TBS buffer was first titrated with MUGB (2 M) prepared in dimethylformamide, and the burst of fluorescence corresponding to the single turnover hydrolysis of MUGB was recorded to generate a standard curve. FXIa mutants (150 nM) were then titrated with MUGB (2 M), and the active site concentrations of the FXIa mutants were calculated from the burst of fluorescence by reference to the standard curve. The active site concentrations of the mutants, expressed as a percentage of the result for either FXIa WT or pFXIa (which gave similar values) were 62-124% (mean ϭ 81%).
Inhibition of FXIa by pAB-pAB is used as a probe to assess the integrity of the S1 site of serine proteases that are argininespecific (26,27) and has been shown to be a reversible inhibitor of serine proteases (28). To test the integrity of the S1 specificity site of the FXIa mutants, pFXIa and mutants were examined for their ability to hydrolyze S-2366 in the presence and absence of pAB. Briefly, either FXIa or one of the mutants (1 nM) was added to wells containing increasing concentrations of pAB (0 -1 mM) and 500 M substrate 2366 in 25 mM Tris, 150 mM NaCl, 5 mM CaCl 2 , pH 7.4. The reaction was performed at 37°C, and the absorbance was read at 405 nm. Inhibition constants were determined by plotting the residual activity of FXIa and by using the following equation, where S is the substrate concentration and K m is the Michaelis constant of FXIa for S-2366 previously determined to be ϳ0.25 mM (29).
Hydrolysis of Substrate 2366-To determine the kinetic parameters of S-2366 hydrolysis by pFXIa and the FXIa mutants, increasing concentrations of S-2366 (0 -1.5 mM) were added to pFXIa or FXIa mutants (6.7 nM final concentration). Linear initial rates of generation of p-nitroaniline (pNA) were measured by continuous monitoring of absorbance at 405 nm in a reaction mixture of 100 l (path length of 3.125 mm) in a microplate reader (Molecular Devices, CA). All our experiments were carried out at a temperature of 37°C in Tris buffers prepared with a pH of 7.4 at room temperature (ϳ25°C). Because the calculated pH at 37°C, based on the ⌬pK a /degree of Ϫ0.031 for Tris and the 12°C temperature difference is 7.0, the actual pH at which our experiments were carried out is likely to be ϳ7.0. Concentrations of pNA were determined using a molar extinction coefficient of 9933 M Ϫ1 cm Ϫ1 . Titration curves of S-2366 hydrolysis were generated by Kaleida-Graph (Abelbeck Software, Reading, PA), and the data were analyzed using a nonlinear least squares fit of data points to the equation for a rectangular hyperbola, y ϭ ax/(b ϩ x).
Activation of FIX by FXIa and FXIa Mutants-To examine the contributions of individual residues clustered around the active site of FXIa to the activation of FIX, either FXIa or one of the FXIa mutants (2.5 nM final concentration) was added to increasing concentrations of FIX (0 -2 M) in 50 mM Tris-HCl with 150 mM NaCl, 5 mM CaCl 2 , and 1% BSA. As described previously (30), the reaction was allowed to proceed for 3 min before it was stopped by the addition of aprotinin (0.7 mM). Ethylene glycol (16%, v/v) was added to increase the sensitivity of the substrate for FIXa before the addition of Spectrozyme (2.6 mM). Linear initial rates of release of pNA were monitored by measuring absorbance at 405 nm at 37°C and analyzed as described previously (30).
Determination of Equilibrium Inhibition Constants for FXIa Mutants by PN2KPI-Increasing concentrations of WT PN2KPI domain (0 -10 nM) in 50 mM Tris-HCl, 150 mM NaCl, 5 mM CaCl 2 , 0.5% BSA, pH 7.3, were incubated with pFXIa or FXIa mutants (1 nM final concentration) for 30 min at 37°C in a microtiter plate to achieve equilibrium between enzyme and inhibitor. Residual FXIa activity in the reaction mixture was determined by hydrolysis of S-2366 (0.5 mM). The absorbance at 405 nm in a microplate reader (Molecular Devices) was measured under pseudo-first-order kinetic conditions, and the results were converted to fraction of amidolytic activity remaining. The concentration at which 50% activity remained (IC 50 ) was determined using Kaleidograph. The inhibition constant was then determined using Equation 1 (see above).
Determination of Inhibition Constants for FXIa Mutants by PN2KPI by Progress Curve Analysis-For those FXIa mutant proteins determined to be deficient in PN2KPI inhibition (i.e. with significant increases in equilibrium inhibition constants), K i values were also determined by progress curve analysis as described previously (31). Briefly, the release of a highly fluorescent product from the peptidyl substrate t-butoxycarbonyl-Glu(o-benzoyl)-Ala-Arg-methylcoumaryl-7-amide by FXIa or mutant molecules was monitored in the presence of varying concentrations of PN2KPI. Substrate hydrolysis was initiated by the addition 10 l of enzyme (25 pM FXIa, final concentration) to 90 l of a mixture of inhibitor (varying concentrations) and fluorogenic substrate in TBSB in 96-well black polystyrene microtiter plates. The progress of substrate hydrolysis was monitored for up to 50 min at 15-or 60-s intervals at room temperature (ranging from 24 to 27°C) in a fluorescence plate reader, and K i values were determined as described earlier (31).
Biotinylation of PN2KPI-PN2KPI was biotinylated using the EZ-Link Micro sulfo-NHS-biotinylation kit obtained from Pierce. The appropriate volume of sulfo-NHS-biotin solution was added to PN2KPI protein solution in phosphate-buffered saline (PBS; 137 mM NaCl, 2.7 mM KCl, 8 mM Na 2 HPO 4 , 1.5 mM KH 2 PO 4 , pH 7.4) and incubated at 4°C for 2 h. Excess biotin was removed by applying the protein solution to a desalting spin column and centrifuging the column at 1000 ϫ g for 2 min. Purified biotinylated PN2KPI was obtained from the collection of the flow-through solution. The concentration of biotinylated PN2KPI was determined by a BCA assay.
Determination of FXIa Binding to PN2KPI by ELISA-100 l of biotinylated PN2KPI (0.5 g/ml) was added to each well of a NeutrAvidin-coated plate (Pierce), which was then incubated at 4°C overnight. Unbound biotinylated PN2KPI was removed, and the wells were washed three times with wash buffer (PBS-Tween, 0.1% (v/v)). 200 l of blocking buffer (PBS-BSA, 1% (w/v)) was added to each well of the plate, which was incubated at room temperature for 2 h. Blocking buffer was then discarded, and the wells were washed three times with wash buffer. 100 l of various concentrations of FXIa WT or FXIa mutants (0 -35 nM) was added into the wells and incubated at room temperature for 30 min. FXIa protein solutions were then discarded. After the wells were washed three times with wash buffer, the detecting antibody, peroxidase-conjugated polyclonal anti-FXI antibody, diluted 100-fold in sample diluent (HBS-BSA-Tween, 100 mM NaCl, 100 mM HEPES, pH 7.2, 1% BSA, and 0.1% Tween 20) was added to each well and incubated at room temperature for 1 h. After the wells were washed thoroughly with wash buffer, 100 l of the substrate, o-phenylenediamine (0.42 mg/ml) in citrate-phosphate buffer (27 mM citrate acid, 97 mM Na 2 HPO 4 , pH 5) was added to each well for 5-10 min for color development. 50 l of stopping solution (2.5 M H 2 SO 4 ) was added to stop the color development reaction. The plate was read at 490 nm using a Thermomax plate reader (Molecular Devices Corp., Sunnyvale, CA).
Molecular Modeling-In the modeling procedure, the energy-minimized light chain of FXIa was compared with the energy-minimized light chain of FXIa with mutations at residues Ile 151 , Glu 98 , Tyr 143 , and Arg 3704 (chymotrypsin numbering). The energy minimization method utilized SYBYL software (version 7.2, from Tripos Inc., St. Louis, MO) with the charge type being AMBER7 F99 and ligands having Gasteiger-Hückel charges. Energies of the models were all minimized by ϳ85,000 kcal/mol. The cut-off point for the calculation was the point at which the force gradient became less than 0.05 kcal mol Ϫ1 Å Ϫ1 or until 10,000 iterations were completed.
Data Analysis-K m , k cat , and K i values for FXIa mutants were compared with the values for pFXIa and analyzed for their statistical significance using an unpaired Student's t test. The Bonferroni adjustment was applied to the p values obtained from the t test in order to correct for a type I error of level 0.05 (an error made by incorrectly declaring an error due to chance). Values were considered statistically significant if p Ͻ 0.05.

RESULTS
Initially we mutated the six residues identified from the crystal structure ( Fig. 1 and Table 1) to alanine (Table 2) and examined their interactions with pAB, S-2366, FIX, and PN2KPI. Some of these residues were further examined for the effects of charge and size of side chains by additional mutations as shown in Table 2.
Inhibition by pAB of FXIa Mutants-The inhibition of FXIa and mutant proteins by pAB was examined in order to assess the integrity of the S1 substrate-binding site of FXIa (27). The results are presented for plasma FXIa and all of the mutant proteins in supplemental Fig. 2, and the derived K i values are summarized in Table 2. Control values throughout the paper are reported for plasma-derived FXIa (referred to as pFXIa), and these values were not significantly different from those obtained with recombinant wild-type FXIa. Therefore, for the sake of conciseness, control values for FXIa WT are not reported. FXIa was inhibited by pAB with an inhibition constant (K i ) of 51.3 Ϯ 1.14 M. All alanine mutants, with the exception of the Y143A mutant (K i ϭ 21.1 M) and the K192A mutant (K i ϭ 152.9 M), displayed K i values ranging from 36.2 to 73.9 M, which were not significantly different from the K i value for pFXIa. The K i value for the E98V mutant (K i ϭ 39.6 M) was also within this range, whereas the K i value for the E98D mutant (K i ϭ 29.6 M) and the K192E mutant (K i ϭ 24.5 M) were only slightly lower, and the K192Q mutant (K i ϭ 79 M) was only slightly higher. We do not regard these minor differences as biologically relevant because inspection of the inhibition data (supplemental Fig. 2) indicate only minor deviations from control curves. Because the K i value for some of these mutants were slightly lower than that of the wild-type protein, suggesting slightly more potent inhibition, these results provide no evidence for a defect in pAB binding to any of the mutant proteins, except possibly minor defects for the K192A and the K192Q mutants. Combined with the results of active site titrations, demonstrating that the mutant proteins retained 62-124% (mean ϭ 81%) of their active site concentration (see "Methods"), these data suggest that the S1 substrate-binding sites of most if not all mutant proteins were largely intact.
Cleavage of Synthetic Substrate S-2366 and FIX by FXIa and FXIa Mutants-The ability of each of the FXIa mutants to cleave the small synthetic substrate, S-2366, was examined at varying substrate concentrations, and the results are presented in supplemental Fig. 3. The K m and k cat values for substrate hydrolysis were determined for each FXIa mutant and are summarized in Table 2 The observation that the K192E mutation had only a minor effect on the k cat of S-2366 hydrolysis compared with FXIa/ K192A, whereas the K192R mutation had a major effect on k cat is counterintuitive and not subject to definitive rational interpretation.
None of the FXIa mutants was able to catalyze the activation of the macromolecular substrate FIX as efficiently as pFXIa (supplemental Fig. 4 and Table 2). Thus, all of the mutants examined displayed 4 -10-fold decreased values of k cat compared with pFXIa (0.73 s Ϫ1 ). As is apparent from inspection of the saturation curves (supplemental Fig. 4), we were unable to calculate reliable values of K m for all of the mutant proteins examined. Thus, all FXIa mutants displayed saturation curves truncated at very low values of V max . For a number of these mutants (e.g. especially for those depicted in supplemental Fig.  4, A (E98D and E98V), B (K192Q and K192R), C (R3704A and Y143A), and D (Y5901V)), due to the insensitivity of the chromogenic substrate (Spectrozyme in the presence of ethylene glycol), the amounts of product (FIXa) generated at low concentrations of FIX are too low to be reliably measured. Therefore, we have not calculated values of K m , and measured values of k cat should be regarded as overestimates. Thus, the k cat values listed in Table 2 most likely underestimate the defects in k cat for many of the mutants studied. For this reason, the values of k cat Ͻ0.2 s Ϫ1 for FIX activation have been listed as such to reflect the insensitivity of the assay and to acknowledge that that the rates were too low to be quantified reliably. Some of the other mutants studied (e.g. supplemental Fig. 4, A (E98A), B (K192A and K192E), C (I151A), and D (Y5901A)) displayed saturation curves truncated at very low values of V max , suggesting decreased values of apparent K m , which typically reflects tighter binding to FIX. It should be noted, however, that the measured K m is not the true substrate dissociation constant (K d or K s ) and that the rate of the reaction contributes to the K m (i.e. K m ϭ K d ϩ k cat /k on ). Therefore, the decreased reaction rates that were observed for some of these FXIa mutants most likely resulted in decreased values of substrate concentration at saturation in the absence of enhanced enzyme-substrate affinity that would be manifested by decreased values of K d . Because these decreased values of K m most likely reflect the defects in k cat rather than enhanced substrate binding, as reported previously (32, 33), we have not listed values of K m for FIX activation in Table 2 and have included a notation that reliable values of K m for FIX activation are not available. The major conclusion to  Table 1.  Table 2 is that the values listed are overestimates, and therefore the defects in k cat are underestimated, thereby strengthening our argument that all of the mutations have deleterious effects on the capacity of FXIa to catalyze FIX activation.

Inhibition of FXIa and FXIa
Mutants by PN2KPI-FXIa and each of the mutant proteins were incubated with various concentrations of PN2KPI (supplemental Fig. 5) to determine K i values, summarized in Table 2. The equilibrium inhibition constant measured for plasma FXIa was 1.5 nM, in close agreement with the K i value determined by progress curve analysis (0.610 nM). As shown in Table 2, mutations at four of these sites (Glu 98 , Tyr 143 , Ile 151 , and Arg 3704 ) resulted in normal values of K i ranging from 1.0 to 1.4 nM, comparable with that of pFXIa, demonstrating that these residues are not essential for inhibition by PN2KPI. In contrast, mutations at the other two sites (Lys 192 and Tyr 5901 ) resulted in significantly increased K i values or no discernible inhibition utilizing the equilibrium inhibition assay. In each instance, these results were confirmed utilizing progress curve analysis, which demonstrated significantly increased K i values ( Table 2). The sole exception to this result was observed with the K192Q mutant, which displayed normal K i values using both methods, suggesting that the presence of an amino group in the side chain of either lysine or glutamine of residue 192 of FXIa is important for inhibition by PN2KPI.
Binding of FXIa and FXIa Mutants to PN2KPI-A possible explanation for the failure of PN2KPI to inhibit FXIa molecules mutated at Tyr 5901 and Lys 192 is the failure of the mutant proteins to bind to PN2PKI. To examine this possibility, we selected four of the FXIa molecules studied here to examine their capacity to bind to PN2KPI in a microtiter plate equilib-rium binding assay. The results are shown in supplemental Fig.  6. Two of these proteins (FXIa wt and FXIa/Y143A) were inhibited with normal K i values (1.5 and 1.3 nM, respectively; see Table 2) by PN2KPI and were shown to bind with K D values (1.63 and 1.97 nM, respectively; see Table 3) very close to the measured K i values. The other two proteins (FXIa/Y5901A and FXIa/K192E) were not inhibited by PN2KPI in the equilibrium binding assay and were characterized by highly significant increases in K i values by progress curve analysis (see Table 2). In the direct binding assay, the FXIa/Y5901A mutant interacted with PN2KPI with normal affinity (K i ϭ 1.22 nM), whereas the FXIa/K192E mutant bound to PN2KPI with a K i value (4.75 nM) only ϳ3-fold higher than FXIa WT . Thus, we conclude that the impaired susceptibility of molecules mutated at Tyr 5901 and Lys 192 to inhibition by PN2KPI arises not from a defect in the binding of KPI to the catalytic domain of FXIa but rather from the fact that these two residues play an essential role in both substrate catalysis and inhibitor recognition.

DISCUSSION
The mechanism by which serine proteases recognize and cleave their substrates and inhibitors involves the formation of a catalytic triad (His 57 , Asp 102 , and Ser 195 ), located at the   Because the architecture of residues located adjacent to the active site is an important determinant of substrate and inhibitor specificity and enzyme activity, we took advantage of the crystal structure of the enzyme-inhibitor complex between the catalytic domain of FXIa and the KPI domain of protease nexin 2 to identify residues potentially important for substrate hydrolysis and/or inhibitor recognition. We also utilized the primary sequence of FXI and a highly homologous protein, PK, to identify residues unique to FXIa, whose macromolecular substrate (i.e. FIX) and major inhibitor (i.e. PN2) are poorly recognized by plasma kallikrein. Finally, residues that are important for enzyme function, including substrate and inhibitor interactions, are characteristically highly conserved among various species (35,36). Employing this combined structural information, we have succeeded in identifying residues that are essential for S-2366 hydrolysis, for interaction with PN2KPI, and for activation of the macromolecular substrate FIX. As demonstrated in Table 2, mutations to alanine of four surface-exposed FXIa residues (Tyr 143 , Ile 151 , Tyr 5901 , and Arg 3704 ) resulted in significant reductions in the rate of S-2366 hydrolysis, whereas effects on k cat by the other two mutants (E98A and K192A) were minimal. In the FIX activation assay, however, mutations at all of the six residues resulted in severely reduced rates of FIXa generation. Therefore, although all six residues were essential for optimal rates of macromolecular substrate (FIX) cleavage to generate FIXa, only four residues (Tyr 143 , Ile 151 , Tyr 5901 , and Arg 3704 ) are involved in hydrolysis of the peptidyl substrate S-2366.
Interestingly, although each of the six FXIa residues chosen for mutational analysis was selected on the basis of its structural interactions with specific KPI residues demonstrated to be important for FXIa inhibition, mutations at four of these sites (Glu 98 , Tyr 143 , Ile 151 , and Arg 3704 ) resulted in normal values of K i (1.0 -1.4 nM), demonstrating that these residues are not essential for inhibition by PN2KPI. In contrast, mutations at the other two sites (Lys 192 and Tyr 5901 ) resulted in significantly increased K i values or no discernible inhibition, demonstrating that residues Lys 192 and Tyr 5901 are essential for interaction with PN2KPI. It is worth emphasizing that whereas three mutants, K192A, K192E, and K192R, showed impaired inhibition, the mutant K192Q inhibited FXIa with K i similar to PN2KPI (Table 2). A plausible explanation is that the presence of an amino group in the side chain of either lysine or glutamine of residue 192 of FXIa is important for inhibition of FXIa by PN2KPI.
It is not surprising that the four mutants Y143A, I151A, Y5901A, and R3704A were found to be deficient in assays of both FIX activation and small peptide hydrolysis because enzyme molecules with impaired catalytic activity against a small tripeptide substrate such as S-2366 would also be expected to have impaired catalytic activity against the normal macromolecular FXIa substrate. However, both E98A and K192A mutants were characterized by only a ϳ53% (insignificant) reduction in the k cat for S-2366 cleavage with a highly significant decrease in FIX substrate catalysis, suggesting that both Glu 98 and Lys 192 are essential for cleavage of the macromolecular substrate FIX. Thus, mutations at either of these two residues may disrupt FIX binding to a FXIa substratebinding exosite (14).
Glu 98 is part of the 90s loop (comprising residues 94 -100) that varies slightly in length and conformation among serine proteases. Comparison of the 90s loop structure of FXIac ( 94 YKMAESG 100 ) ( Fig. 2A) with that of PK ( 94 YKVSEGN 100 )  2ANY; B). A, some of the residues that were mutated (Arg 3704 , Glu 98 , Tyr 143 , and Ile 151 ) are highlighted, and the catalytic triad is depicted in red. The autolysis loop (residues 143-154) of FXIac is colored gray. B, close-up view (in the same orientation as in A) of the 90s loop (residues 94 -100) of FXIa and kallikrein. The FXIa backbone is shown in light blue, whereas kallikrein is shown in green. The catalytic triad residues for FXIa are red, and those for PK are yellow. The 90s loop, including residue Glu 98 , which is shown in a stick model format, is gray for FXIa and purple for kallikrein. (Fig. 2B) indicates that in FXIa, the 90s loop folds inward toward the catalytic triad residues and causes this region to be more restricted than the comparable region of PK (Fig. 2B). Both FXIa and kallikrein have a Glu at position 98; however, because the conformation of the 90s loop is different, the side chains are positioned differently (Fig. 3A). In FXIa-benzamidine, rhFXIa-(370 -607)-ecotinM84R, and FXIa-PN2KPI structures, the aliphatic portion of the side chain of Glu 98 forms a van der Waals interaction with Trp 215 (21, 37, 38) (see Fig. 3A) is within hydrogen bonding distance of His 174 (3.96 Å). These interactions restrict the region and block solvent accessibility to the indole side chain. Mutation of Glu 98 to Asp or Val disrupted these interactions and adversely affected the catalytic function of FXIa; thus, compared with pFXIa, the k cat values for S-2366 hydrolysis by the E98D and E98V mutants were reduced by ϳ35-fold and ϳ10-fold, respectively. The fact that the mutation of Glu 98 to an alanine resulted in a minor (2-fold) decrease in k cat against S-2366, whereas all three Glu 98 mutants caused a highly significant decrease in k cat against FIX suggests that residue Glu 98 may be important for macromolecular substrate selectivity. Thus, the striking differences between FXIa and kallikrein in the positions of Glu 98 (Figs. 2B and 3A) and His 174 (Fig. 3A) most likely account partially for their striking differences in catalytic activity and explain the essential role of Glu 98 in catalysis of the macromolecular substrate, FIX.
Our molecular modeling studies also predict perturbations of the interactions of neighboring amino acids due to mutations at residues identified herein to be important for substrate hydrolysis but not inhibitor interaction. The positions of the various mutants at positions Arg 3704 , Glu 98 , Tyr 143 , and Ile 151 with respect to active site residues (Ser 195 , His 57 , and Asp 102 ) are shown in Fig. 4A along with Ser 214 , Trp 215 , Asp 194 , and the NH 2 -terminal residue Ile 16 . These interactions are important for determining the architecture of the FXIa active site, which is perturbed when mutations of some of these crucial residues are introduced, resulting in impaired catalysis. Glu 98 lies within 4 Å of Trp 215 , with which it appears to have a weak van der Waals interaction. The function of the Ser 214 residue is not fully accounted for. In thrombin, mutation of Ser 214 to alanine caused a loss in activity (39), whereas in trypsin, a significant increase in activity occurred (40). The backbone of Ser 214 is known to be important in substrate binding because it forms a hydrogen bond with the P1 residue of the substrate (41). Mutations of Glu 98 in the model (Fig. 4B) showed movement of the relative position of the ␣-carbon of the mutated residue and the Trp 215 ring and the relative position of Ser 214 , the hydroxyl group of which forms a hydrogen bond with the carboxyl oxygen of the Asp 102 . Although it is difficult to predict from the model the extent of perturbation of the architecture of the active site by different residues at position 98, it is reasonable to assume that it is dependent on the size as well as charge of the mutated residue. As shown in Table 2, the E98D mutant showed the largest decrease in amidolytic activity (ϳ35-fold), valine showed a slightly lower decrease (ϳ10-fold), and the alanine mutant displayed a statistically insignificant decrease (ϳ2fold), whereas all these mutants manifested major decreases in k cat for FIX activation. Although all of the residues mutated are on the surface of the molecule, serine proteases are very sensitive to even small changes. This is demonstrated by their large presence in biological systems, as they comprise the largest protease family, mainly resulting from mutations in only ϳ50 residues. This illustrates the wide range of functions and unique interactions they are able to perform from a shared structure from only small variations (41,42).
The autolysis loop (Tyr 143 -Thr 154 ) is a surface-exposed loop structure that has previously been shown to be important for substrate and inhibitor specificity in FIXa (43), FXa (44), FXIa (45), thrombin (46), urokinase-type plasminogen activator (47), and activated protein C (48). In FIXa, residues Arg 143 and Lys 147 have been shown to be involved in substrate (FX) recognition, and Arg 150 has been shown to be involved in interaction with antithrombin when in the heparin-activated conforma- tion (43). Residues Tyr 143 and Ile 151 of FXIa are both contained within this loop (Fig. 1A). Tyr 143 is not conserved among serine proteases; however, in all species for which the FXI sequence is known, Tyr at position 143 is conserved. In the species for which the FXI sequence is known, residue 151 is typically an Ile or a Val. In Homo sapiens, residue 151 is an Ile. Based on the crystal structure of FXIa in complex with PN2KPI (21), Tyr 143 of FXIa forms a van der Waals interaction with Met 17 of PN2KPI, and Ile 151 is within close proximity to Met 17 . We therefore anticipated the possibility that mutation of Tyr 143 and Ile 151 to alanine might impair the interaction of the resulting mutants with PN2KPI. However, in experiments measuring the residual activity of Y143A after preincubation with PN2KPI, there was no observed difference in the inhibition compared with that of pFXIa. In assays measuring amidolytic activity, however, the Y143A mutant demonstrated a normal K m ; however, the k cat value was decreased ϳ30-fold compared with pFXIa. Likewise, Y143A was defective in catalyzing FIX activa-tion (Table 2). Likewise, the K m for hydrolysis of S-2366 by FXIa I151A was not significantly different from that for pFXIa. Catalysis of S-2366, however, was affected by this mutation with a ϳ3.5-fold decrease in k cat value compared with pFXIa. Similarly, the k cat value for catalysis of FIX was decreased ( Table 2). The K i value for the inhibition of Y143A or I151A by pAB was not significantly different from that for pFXIa, indicating that there was no gross structural rearrangement of the S1 specificity site. Tyr 143 is within hydrogen bonding distance to Lys 192 (Fig. 3B), a residue that has been demonstrated to be important for inhibitor and substrate specificity in trypsin, FXa, activated protein C, and thrombin (49 -53). Also within close proximity is Ile 16 , the residue that forms a salt bridge with Asp 194 upon activation of FXI (Fig. 4A). By changing Tyr 143 or Ile 151 to Ala, the interaction of residue 143 with Lys 192 may be adversely affected, thereby causing the overall structure of the oxyanion hole, formed by the adjacent residues Gly 193 and Ser 195 , to be affected as well. The amide nitrogen of Gly 193 may no longer be properly oriented, thereby not allowing the complete formation of the oxyanion hole. From these experiments, we conclude that Tyr 143 and Ile 151 have no important role in inhibition by PN2KPI but are important for the normal catalysis of both FIX and S-2366.
In our molecular modeling studies (Fig. 4C), no major conformational alteration in the interaction of Ile 16 with Asp 194 for FXIa/Y143A or FXIa/I151A could be demonstrated. However, a significant increase in energy (35.4 kcal/mol as a consequence of the Y143A mutation and 3.46 kcal/mol as a consequence of the I151A mutation) was observed in the autolysis loop (Tyr 143 -Thr 154 ), reflecting disordered structure. Upon inspection of the crystal structure of zymogen FXI (54), it was found that the autolysis loop contains a high amount of thermal movement, as indicated by b-factors of 140 -143 in Lys 145 and 103-117 in Gly 142 , whereas in contrast, in the crystal structure of the enzyme FXIa (21), the b-factors of Lys 145 (79 -83) and Gly 142 (63-65) were significantly lower. This suggests that the movement of the N-terminal amine of the Ile 16 residue into close proximity (2.8 Å) of the backbone oxygen of Tyr 143 has a stabilizing effect upon the autolysis loop. The change in energy effected by the I151A mutation is postulated to result in altered dynamics within the autolysis loop before Ile 151 has a chance to interact with and stabilize it. Because it is known that the autolysis loop undergoes a significant movement after activation in serine proteases and because it is in the activation domain (41), the mutation of Ile 151 to alanine may reduce the likelihood of successful active site formation and the resultant impaired activity of the mutant enzyme.
Previous studies focusing on the contributions of basic residues within the autolysis loop of FXIa to serpin specificity (45) determined that mutagenesis of basic residues within the autolysis loop (Arg 144 , Lys 145 , Arg 147 , and Arg 149 ) has no adverse effects on the conformation of the FXIa S1-S3 substrate-binding sites or the capacity of these autolysis loop mutants to cleave the chromogenic substrate S-2366. Our present studies demonstrate that mutation of two additional residues within the autolysis loop of FXIa, Tyr 143 and Ile 151 , have no adverse effects on the conformation of the FXIa S1-S3 substrate-binding sites or the PN2KPI binding site. However, these two residues are required for normal substrate hydrolysis, as shown in Table 2 and depicted in Figs. 1 and 4A, in which a hydrophobic interaction between Tyr 143 and Ile 151 is demonstrated to be part of interactions involving Lys 192 , Asp 194 , and Ile 16 that are essential for maintaining the normal architecture of the catalytic site.
Arg 3704 is a residue unique to FXI and is located on a surfaceexposed loop region (residues 36 -3704, chymotrypsin numbering, or 390 -395 in mature FXI or 71-76 of the catalytic domain of FXIa) that is slightly longer than that in other serine proteases. We investigated this unique residue to assess its importance in FXIa activity. The S1 site of the R3704A mutant was intact, with a K i value for pAB binding comparable with that of pFXIa. The value of k cat was significantly decreased for S-2366 cleavage (ϳ25-fold) compared with pFXIa and was also decreased in FIX catalysis (Table 2). However, it is unclear from inspection of the FXIa crystal structure (Figs. 1 and 4D) exactly what contribution Arg 3704 makes to substrate hydrolysis. From the FXIa-PN2KPI structure ( Fig. 1 and Table 1), we observed that Arg 3704 is within hydrogen bonding distance with the main-chain nitrogen atom of Met 17 and the O␥ of Ser 19 of PN2 (21). Despite the proximity of both Met 17 and Ser 19 to Arg 3704 , a loss of inhibition of this mutant by PN2KPI was not observed. Experiments with alanine substitutions at Met 17 and Ser 19 of PN2KPI showed a minimal loss of inhibitory function toward pFXIa with a 1.3-and 1.5-fold decrease in activity, respectively (21). In our molecular modeling studies, the mutation of the Arg 3704 residue to alanine had a significant effect on the local backbone position (Fig. 4D), which was accompanied by a 4.3kcal/mol rise in energy associated with the surrounding loop structure. This is probably a consequence of an alteration in the interaction between the side chains of Ile 151 and His 38 . However, unlike the small decrease in activity that arises from the I151A mutation, the R3704A mutant displayed a large (25-fold) decrease in activity. Therefore, there is likely to be another mechanism accounting for the relatively large effect of the R3704A mutation that is not displayed in the model.
In summary, based on the structure of the complex between the catalytic domain of FXIa and the KPI domain of PN2, we have identified six residues clustered around the active site of FXIa for mutational analysis. All 12 mutant molecules prepared were characterized by reasonably normal S1 specificity site binding to pAB and normal binding affinity to the small peptide substrate S-2366, suggesting normal active site architecture. Mutations at four sites (Glu 98 , Tyr 143 , Ile 151 , and Arg 3704 ) produced enzymes that were normal in inhibitor (PN2KPI) recognition, whereas mutations at the two remaining sites (Lys 192 and Tyr 5901 ) produced enzymes that were deficient in inhibitor (PN2KPI) recognition. In addition, mutations at all six residues (Glu 98 , Tyr 143 , Ile 151 , Arg 3704 , Lys 192 , and Tyr 5901 ) resulted in enzymes that were deficient in macromolecular (FIX) substrate catalysis. Of the six residues, only four (Tyr 143 , Ile 151 , Arg 3704 , and Tyr 5901 ) were found to be important for small substrate hydrolysis. Impaired cleavage of macromolecular substrate (FIX) and not of the small peptide (S-2366) hydrolysis by the two mutants FXIa/E98A and FXIa/K192A indicates that mutations at these two residues (Glu 98 and Lys 192 ) disrupt exosite interactions of the enzyme with FIX, suggesting that these two residues are part of an exosite for macromolecular (FIX) but not small peptide substrate activation.